Offline Multi-Agent Reinforcement Learning with Knowledge Distillation